Recipe Optimization for Post-CMP Defect Inspection
Liang Shi and Leo Veltman Philips Semiconductors Nijmegen, Netherlands Brian Zhang and Ulrich Winkler Applied Materials, Santa Clara, Calif. Frank Verstraete and Robert Schreutelkamp Applied Materials, Nijmegen, Netherlands -- Semiconductor International, 7/1/2001
| At a Glance |  | ||
| | ||
| |||
For wafer defect inspection, a key requirement is to ensure a consistent and methodical recipe setup procedure that will lead to high-sensitivity and robust inspection recipes for current layer defect inspections, such as the post-CMP inspections on patterned wafers. In many instances, recipe sensitivity of patterned wafer inspection is evaluated by the actual defects found on the setup wafers.
Depending on process variations and deviations over time, the number of defects, defect types and defect sizes on the wafers can vary significantly. Without a consistent reference media, it is difficult to optimize and compare recipe sensitivity. Test wafers with programmed defects are useful for checking the general conditions of inspection systems, but they do not in any way resemble production wafer conditions and thus cannot provide any guidance to recipe optimization of production wafers.
Robustness has frequently been ignored in setting up production inspection recipes. One of the parameters to measure robustness is defect capture rate. Since production wafers are inspected only once at any inspection step, it is critical that a high capture rate is obtained for the defects of interest. Unfortunately, defects on production wafers do not have uniform size, shape and composition to allow for meaningful statistical analysis of defect capture rate. Therefore, no information about defect capture rate can be obtained practically from production wafers.
PSL spheres have been widely used for recipe optimization for blanket/ bare wafer inspection. These wafers offer a highly uniform and low-noise background to allow for inspection systems to easily separate defects from background. Because of the highly accurate sizing and size uniformity of the spheres, sensitivity of inspection systems can easily be derived from scanning results and a response (to different sizes of PSL spheres) can be generated to optimize the recipe accordingly for sensitivity.
The application of PSL spheres on patterned wafers for recipe optimization purposes is, on the other hand, relatively unknown. By utilizing PSL spheres on wafers of different process steps, it is possible to directly compare the sensitivity of recipes, regardless of the background noise.
In particular, since the PSL spheres are surface defects, optimized inspection recipes can be directly applied to CMP inspections where the defects of interest are all near-surface. Even though Applied Materials' brightfield/darkfield patterned wafer inspection tool has been successfully used for tungsten (W) CMP inspections at Philips MOS4, 1 it has been demonstrated that the use of the PSL spheres has been instrumental in optimization of W CMP inspection recipes.2
Materials and preparation
 1. This map shows the distribution of PSL spheres on a wafer. The A, B, C and D regions correspond to PSL sphere sizes 0.198, 0.304, 0.451 and 0.966 µm, respectively. |  |
For this study, two logic production wafers from Philips' MOS4 wafer fab are used (0.25 µm design rule and fully processed until after via 1 W CMP). Each wafer was deposited with four different sizes of PSL spheres on four different locations on the wafer. The deposition was done in Applied
Materials' Defect and Thin Film Characterization Lab in Santa Clara, Calif. Scanning electron microscope (SEM) images were taken from Applied Materials' SEMVision CX. Figure 1 highlights the distribution of the PSL spheres on the wafers.
Figure 2 shows images of a 0.198 µm PSL sphere on one of the post W CMP production wafers under CCD camera on the inspection tool in darkfield mode and a SEMVision CX image of the same 0.198 µm PSL sphere. The accurate and uniform sizing of the PSL spheres, evident from the SEM image, is critical for conducting statistical analysis such as estimating the capture rate as a function of defect size.
A recipe was first set up on the wafer inspection tool at Applied Materials' application laboratory based on a best-known method (BKM) as reference. To detect surface defects such as microscratch and slurry residue, two critical conditions have to be met: optimal focus on the surface of the wafer and optimal sensitivity to surface defects. The first condition is achieved by selecting an area of interest on the current layer and the wafer inspection tool can automatically adjust its focus offset to find the optimal focus to the surface of the wafer. The achievement of the second condition is greatly facilitated with the use of PSL spheres.
|  2. SEMVision image (left) and darkfield CCD image (right) of a 0.198 µm PSL sphere on a post W CMP production wafer. |
Even though the optimal focus can be achieved without PSL spheres, it is difficult to optimize detection sensitivity to surface defects without an unambiguous reference on the wafer surface. Overall sensitivity was optimized such that spheres for each size were detected. Subsequent study on the capture rate of the PSL spheres would give further indication to the sensitivity and robustness of the inspection recipe.
Lab results
Each of the two wafers was scanned 10 times, loading and unloading the wafer between each scan. A defect count repeatability can be calculated as
R=1-s/µ where R is the defect count repeatability, µ is the mean of defect count from the 10 repeats and s is the sample standard deviation. For both wafers, this value is close to 97%, indicating that there is a high repeatability.
For each wafer, the 10 repeat results were overlaid to obtain a union defect map. Defect search radius for the overlay was set at 50 µm. The union defect map contains all the defects detected on the wafer during any of the 10 repeat scans.
The union defects were then manually reviewed and classified in such a way that individual defect files could be created for each type of process defect as well as each size of PSL sphere. To calculate the capture rate of a certain PSL sphere size, each of the 10 repeat scan results was compared to its corresponding defect file, with a coordinate search radius again set at 50 µm.
It is without question that a higher number of repeat scans will lead to a more accurate estimate of the capture rate of defects. In a separate study on different patterned wafers, however, it was found that nearly 90% of the union defects were detected in any 10 repeats of the 30 repeat scans.
Naturally, defect capture rate itself determines how many repeats are sufficient. For example, a defect with a capture rate of 0.2 will have 90% probability to be detected in any of the 10 repeat scans. Therefore, an inspection of 10 repeats is adequate for practical purposes of estimating defect capture rate.
| 3. Capture rate of PSL spheres vs. sphere size on a production wafer, based on 10 repeat scans on Applied Materials’ brightfield/darkfield inspection tool. |
Figure 3 indicates that both good sensitivity (detection down to PSL size of 0.198 µm) and high capture rate (³ 50% for each considered PSL size) are achieved on this production wafer. Similar results were obtained for the other production wafer. It should be mentioned here that these W CMP wafers showed very high background noise from the previous layer structures.
As discussed earlier, defect capture rate cannot be easily obtained from production wafers alone because of the variation in size, shape and composition of the process defects. However, with the use of PSL spheres, both sensitivity and capture rate can easily be obtained on a single production wafer. Furthermore, this is achieved at nearly negligible cost compared with the manufacturing of programmed defect wafers.
Production results
| 4. Comparison of defect types between standard and optimized production recipes. Defect types are based on expert manual classification. |
A comparison of defect count and defect count repeatability showed excellent consistency(<5% difference) between the inspections at MOS4 and at Applied Materials Santa Clara on these two production wafers containing PSL spheres. An overlay (with proper rotation of one of the results to match notch orientation) with 50 µm search radius matched more than 60% of the defects by defect coordinates.
It is important to note here that this was achieved without any hardware matching of the inspection systems, and with essentially the same recipe but different notch orientation and slightly different masking. It can be anticipated that the overlay matching would be much higher had the same orientation and masking been used.
This is a quite significant result since this methodology would allow (virtual) wafer fabs in which similar inspection systems are not yet matched in terms of hardware to conduct essentially the same defect inspection for similar processes.
Production verification
To test the robustness and transferability of this methodology, a similar (optimized) recipe was created for a different product at Philips MOS4. An extensive manual review of several production lots inspected with both the standard and the optimized recipe showed that the optimized recipe detected a much larger number of (micro-)scratches and other small and large defects relevant to W CMP processing compared with the standard recipe.
| 5. Brightfield CCD images of typical defects detected with optimized recipe for a production lot. |
Figure 5 and Figure 6 show some typical defects detected from the production lot with the optimized recipe. Without such methodical approach to recipe optimization at post W CMP layers, most of the scratches would have been missed with the standard defect inspection recipe.
Conclusion
PSL spheres have been successfully used to develop a methodology of recipe optimization for post W CMP defect inspections. The PSL spheres on the wafer surface provide reference points for developing surface-sensitive recipes.
The uniformity of PSL sphere size, shape and composition as well as their ready availability allow the quantitative measurement of sensitivity of recipe and capture rate of the PSL spheres, both of which are critical for developing sensitive and robust inspection recipes.
| 6. Typical SEM images of microscratch defects. |
Phone: +31 24 353 45 21
e-mail: liang.shi@philips.com
Leo Veltmanis an engineer in the metrology and defectivity reduction section of MOS4YOU at Philips Semiconductors. Previously, he worked in product and process development at discrete semiconductors at Philips Stadskanaal. He received his M.Sc. in chemical technology from the Twente University of Technology.
Brian Zhang is a member of the technical staff with the methodical defect reduction group of Applied Materials. He has also been an application development engineer for WF systems with the process diagnostics and control division. Prior to joining Applied Materials, he was a senior process engineer at Intel. He received an M.S. and Ph.D. in materials science and engineering from Purdue University, where he spent one year as a post-doc in the field of high-temperature superconductors.
Ulrich Winkler is a senior director for methodical defect reduction engineering at Applied Materials. Prior to joining Applied Materials, he held various senior management positions in the semiconductor equipment industry. He received an M.S. in physics from the University of Dortmund.
Frank Verstraete is a process diagnostics and control technologist for Applied Materials' global account for Philips. He previously held application and product marketing responsibilities in the division. Prior to joining Applied Materials, he held several process engineering positions at Alcatel Microelectronics and Micronas Intermetall. He received his B.Sc. in chemistry from the Industriele Hogeschool.
Robert Schreutelkamp is a methodical defectivity reduction technologist for Applied Materials' global account for Philips. He previously held application and product marketing responsibilities in Applied Materials' process diagnostics and control division. Prior to joining Applied Materials, he held several technology positions at Alcatel Microelectronics and IMEC. He received his M.Sc. and Ph.D. in physics from the University of Utrecht.
REFERENCES
- L. Shi, et al, "Process Monitoring at Philips' MOS4YOU," European Semiconductor, March 2000, p. 27.
- L. Shi, et al, "Defect Detection at W CMP Layers: Challenges and Successes With WF-736DUO," to be published.
The authors acknowledge Venkat Nagaswami for his valuable input in preparing this paper.
